Yeah, this is one of the weaker parts of the post because I didn't really make my position clear: that sandboxing should be a first-class integral part of every harness, not an afterthought that risks putting something you care about on the same side of the trust boundary as the LLM.
Some of my own frustrations with bad Docker deployments leaked through. I did snip some of them out way back when I posted this, but maybe not enough.
> that sandboxing should be a first-class integral part of every harness,
Essentially, what you want is that the people who produce harnesses like OpenCode to also implement proper sandboxing and security. Why... why do you believe they'll do anything remotely satisfactory?
It's been empirically demonstrated, again and again, that at least 85% of software developers (and this is a very generous estimation) can not be trusted to bother about the security, so leaving the implementation of the security mechanisms and policies up to them simply doesn't work in practice. It just doesn't.